Biology subjects

Seth, S.

Publications and source records attributed to Seth, S..

16 recordsLinked to original sources

bioRxiv2026

CGAgentX: Agentic AI Framework to Develop Transferable Coarse-Grained Models

We present CGAgentX, a general autonomous multi-agent framework in which specialized LLM-based agents coordinate the optimization of coarse-grained (CG) model parameters to reproduce target properties. Using polar solvents -- dimethyl sulfoxide (DMSO) and N,N-dimethylacetamide (DMA) -- as representative case studies, we demonstrate the frameworks capability to develop CG models that accurately reproduce key properties from atomistic simulations and experimental literature. Six specialized agents -- Mapping, Topology, Boundary, Hypothesis, Diagnostic, and Optimization -- operate under a Master Agent that orchestrates closed-loop, iterative parameter refinement by autonomously invoking external tools, including molecular dynamics (MD) simulations and analysis workflows, and evaluating outputs through a fitness function. Central to the framework is a Hypothesis Agent that generates and verifies physically motivated parameter hypotheses by coordinating parallel multi-fork simulations, wherein multiple candidate parameter sets are evaluated simultaneously. This multi-fork strategy expands parameter space exploration, yielding richer datasets that enable more accurate hypothesis refinement across iterations. Agents adaptively propose parameter updates based on intermediate simulation outcomes, enabling efficient navigation of complex trade-offs among structural, thermodynamic, and transport properties. The framework reproduces key experimental properties within 5% accuracy while maintaining consistency with atomistic reference behavior, achieving convergence without manual intervention. The modular architecture is readily extensible to other molecular systems and can accommodate additional targets, constraints, or simulation engines, providing a general agentic-AI platform for transferable CG model development. TOC GRAPHICS O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=110 SRC="FIGDIR/small/719081v1_ufig1.gif" ALT="Figure 1"> View larger version (48K): org.highwire.dtl.DTLVardef@1caa787org.highwire.dtl.DTLVardef@1bcb5eaorg.highwire.dtl.DTLVardef@4b21d2org.highwire.dtl.DTLVardef@999675_HPS_FORMAT_FIGEXP M_FIG C_FIG

biophysics↗

bioRxiv2025

Prediction of physical characteristics of disordered proteins using molecular simulation and physics-informed multiple machine learning strategies

We introduce a novel hybrid machine learning (ML) framework to predict the radius of gyration and other conformational properties of intrinsically disordered proteins (IDPs). Our model integrates sequence information with physical features derived from a coarse-grained (CG) model validated by experimental data. Specifically, we combine hidden states from sequence-based models with 23 physical features projected into a shared latent space, and apply an attention mechanism that assigns weights to each residue to highlight the most informative regions of the sequence. This attention-guided fusion significantly improves predictive accuracy across multiple metrics, including MAPE and MSE, while also enhancing confidence in the predictions. We trained and evaluated our models on Brownian dynamics (BD) simulation results for approximately 7,000 IDPs from the MobiDB database (each with > 99% disorder score). We find that sequence-based models consistently outperform feature-only models, with the GRU achieving the best performance among sequence-only approaches. Moreover, combining sequence and feature information further improves accuracy across all architectures, with the hybrid biGRU model delivering the best overall predictive performance. SHAP analysis reveals the relative importance of physical features, offering model explainability and guiding feature selection. Notably, using a small number of top features often reduces model complexity and improves generalization. Furthermore an integrated gradient analysis reveals that apart from the length of the IDPs, the three parameters (SCD, SHD, and f *) play key role in ML predictions. Our framework provides a fast, interpretable, and scalable tool for predicting IDP behavior, enabling efficient initial screening prior to costly molecular simulations.

biophysics↗

bioRxiv2025

Integrative plasma lipidomics and proteomics profiling to decipher potential biomarkers of dilated cardiomyopathy

BackgroundDilated cardiomyopathy (DCM), primarily characterised by left ventricular dilatation and systolic dysfunction, is one of the leading causes of heart failure and requires a critical clinical investigative strategy. However, conventional imaging techniques such as echocardiography and MRI, along with some classical CVD markers (NTproBNP, cTnT), fall short in diagnosing DCM-specific phenotypes. Thus, the need for biochemical markers with enhanced accuracy to DCM is of enormous importance. Lipids and proteins play essential roles in maintaining myocardial function. The homeostatic disruption of such biomolecules might contribute to DCM pathogenesis, thus offering to serve as potential biomarkers for DCM. Moreover, the lack of global lipidomics studies and specific protein markers in DCM patients prompted us to explore the disease pathophysiology through an integrative omics-based analysis coupled with an ML-derived approach. ObjectiveTo identify accurate, precise and specific circulatory lipidomic and proteomic biomarkers of DCM using high-resolution mass spectrometry and machine learning (ML)-based approaches. MethodsHigh-resolution-mass-spectrometry-based lipidomics and proteomics were applied to identify lipid and protein biomarkers in a cohort (n=360) of healthy and DCM patients. Top protein classifiers were further evaluated using single-cell transcriptomics on publicly available datasets from DCM myocardium and validated using ELISA. A biomarker panel was built by the integration of lipidomics and validated proteomics data using machine-learning-based approaches. ResultsA total of 125 lipids and 10 proteins have been primarily discovered with significant alterations in DCM (0.8 [≥] FC [≥] 1.2; padj < 0.05). Using a Boruta-based ML approach, we identified 39 lipids and 10 proteins as primary discriminators between DCM and controls. ELISA validation confirmed the potential of B2M (6.85 {+/-} 2.86 g/ml vs. 4.26 {+/-} 1.25 g/ml; p < 0.0001) and Tetranectin (CLEC3B) (1.99 {+/-} 0.88 g/ml vs. 2.49 {+/-} 0.90 g/ml; p = 0.0006) to emerge as protein biomarkers of DCM. In line with that, the single-cell transcriptomic analysis showed a similar trend of alteration of tetranectin (CLEC3B) in cardiomyocytes and {beta}2microglobulin (B2M) in varied cell types of the myocardium. Further, Integrated ROC analysis combining the top 8 lipid discriminators with B2M and CLEC3B achieved an AUC of 0.995, demonstrating enhanced diagnostic precision compared to the classical CVD marker NTproBNP (0.965). ConclusionThis study offers a system-omics-based perspective on first-global lipidomic and proteomic changes associated with DCM pathophysiology, with a high potential for diagnostic application.

systems biology↗

bioRxiv2025

Evolution of chronic lymphocytic leukemia after allogeneic stem cell transplant

Allogeneic stem cell transplant (alloSCT) for patients with relapsed/refractory chronic lymphocytic leukemia (CLL) can result in cure in some patients. Analogous to chemotherapy and targeted therapy, we hypothesized that allogeneic cellular immunotherapies, including alloSCT and donor lymphocyte infusion (DLI), would impact malignant evolution through the application of selective immunologic pressure with reciprocal changes in the T cell compartment. We tested a cohort of 24 patients treated with HLA-matched alloSCT +/- DLI, two mediators of the graft versus leukemia (GVL) effect. Comparison of pre-alloSCT samples revealed that a key difference between responders (n=13) and non-responders (n=11) is the cellularity of leukemic cells. We further mapped mutational trajectories of tumor cells by whole exome sequencing (WES) of sort-purified CLL in 11 post-transplant relapsed patients and found evidence of subclonal leukemic evolution in 8/11 patients after nonmyeloablative human leukocyte antigen (HLA)-matched alloSCT. Different patterns of CLL evolution were observed, and these changes included putative CLL drivers in every case. To investigate the presence of immune-related variants in patients, we collected 19 T cell co-culture CRISPR datasets and identified the top positive and negative regulators of cancer cells response to T-cell-dependent killing. We found that most mutations linked to T-cell killing emerged after allo-SCT treatment, suggesting that selective pressures from the GVL effect may drive the evolution of these mutations. Together, these data identify cellular homogeneity as a key biomarker for susceptibility to GVL-driven immunity in CLL and illustrate how the leukemic cells further evolve to evade immunosurveillance. SIGNIFICANCEThe impact of allogeneic stem cell transplant (alloSCT) on subclonal leukemia evolution remains poorly understood. By performing whole exome sequencing (WES) on pre- and post-alloSCT patient samples, we reveal different patterns of CLL evolution and potential immune-related driver genes influencing CLL relapse and refractory disease post-alloSCT.

cancer biology↗

bioRxiv2024

Accelerated Missense Mutation Identification in Intrinsically Disordered Proteins using Deep Learning

We use a combination of Brownian dynamics (BD) simulation results and Deep Learning (DL) strategies for rapid identification of large structural changes caused by missense mutations in intrinsically disordered proteins (IDPs). We used [~] 6500 IDP sequences from MobiDB database of length 20 - 300 to obtain gyration radii from BD simulation on a coarse-grained single bead amino acid model (HPS2 model) used by us and others [Seth et al. J. Chem. Phys. 160, 014902 (2024), Dignon et al. PLOS Comp. Biology, 14, 2018, Tesei et al. PNAS, 118, 2021] to generate the training sets for the DL algorithm. Using the gyration radii [<]Rg[>] of the simulated IDPs as the training set, we develop a multilayer perceptron neural net (NN) architecture that predicts the gyration radii of 33 IDPs previously studied using BD simulation with 97% accuracy from the sequence and the corresponding parameters from the HPS model. We now utilize this NN to predict gyration radii of every permutation of missense mutations in IDPs. Our approach successfully identifies mutation-prone regions that induce significant alterations in the radius of gyration when compared to the wild-type IDP sequence. We further validate the prediction by running BD simulations on the subset of identified mutants. The neural network yields a (104 - 106)-fold faster computation in the search space for potentially harmful mutations. Our findings have substantial implications for rapid identification and understanding diseases related to missense mutations in IDPs and for the development of potential therapeutic interventions. The method can be extended to accurate predictions of other mutation effects in disordered proteins.

bioinformatics↗

bioRxiv2023

Understanding the molecular basis for enhanced glutenase activity of actinidin

Management of gluten intolerance is currently possible only by consumption of gluten free diet(GFD) for a lifetime. The scientific community has been searching for alternatives to GFD, like inclusion of natural proteases with meals or pre-treatment of gluten containing foods with glutenases. Actinidin from kiwifruit has shown considerable promise in digesting immunogenic gliadin peptides as compared to other plant derived cysteine proteases. Through this article, we have attempted to understand the structural basis for elevated protease action of actinidin against gliadin peptides by using an in silico approach. Docking experiments reveal key differences between the binding of gliadin peptide to actinidin and papain, which may be responsible for their differential digestive action. Sequence comparison of different plant cysteine proteases highlights amino acid residues surrounding the active site pocket of actinidin that are unique to this molecule and hence likely to contribute to its digestive properties. Graphical summary O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=64 SRC="FIGDIR/small/542047v1_ufig1.gif" ALT="Figure 1"> View larger version (25K): org.highwire.dtl.DTLVardef@4fcd61org.highwire.dtl.DTLVardef@134c371org.highwire.dtl.DTLVardef@10b5d78org.highwire.dtl.DTLVardef@1257d19_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2021

GlyNet: A Multi-Task Neural Network for Predicting Protein-Glycan Interactions

Advances in diagnostics, therapeutics, vaccines, transfusion, and organ transplantation build on a fundamental understanding of glycan-protein interactions. To aid this, we developed GlyNet, a model that accurately predicts interactions (relative binding strengths) between mammalian glycans and 352 glycan-binding proteins, many at multiple concentrations. For each glycan input, our model produces 1257 outputs, each representing the relative interaction strength between the input glycan and a particular protein sample. GlyNet learns these continuous values using relative fluorescence units (RFUs) measured on 599 glycans in the Consortium for Functional Glycomics glycan arrays and extrapolates these to RFUs from additional, untested glycans. GlyNets output of continuous values provides more detailed results than classification models. Such continuous outputs are easily converted by a following classifier, and in this form GlyNet outperforms reported classifiers. GlyNet is the first multi-output regression model for protein-glycan interactions and will serve as an important benchmark, facilitating development of quantitative computational glycobiology.

bioinformatics↗

bioRxiv2020

Medium-chain acyl-CoA dehydrogenase, a gatekeeper of mitochondrial function in glioblastoma multiforme

Glioblastoma (GBM) is among the deadliest of human cancers. Despite extensive efforts, it has proven to be highly resistant to chemo- and immune-based therapeutic strategies, and little headway has been made with targeted inhibitors. Like many cancers, metabolism is dysregulated in GBM. Thus, to identify new vulnerabilities and drug targets in GBM, we conducted genetic screens using pooled RNAi libraries targeting metabolic enzymes. We screened multiple glioma stem cell-derived (GSC) xenograft models, which revealed that several enzymes involved in the mitochondrial metabolism of fatty acids were required for tumor cell proliferation. From among these, we focused on medium-chain acyl-CoA dehydrogenase (MCAD), which oxidizes medium-chain fatty acids, due to its consistently high score across all of our screens, as well as its high expression level in multiple GSC models and its upregulation in GBM compared to normal brain. In this manuscript, we describe the dependence of GBM on sustained fatty acid metabolism to actively catabolize lipid species that would otherwise damage the mitochondrial structure. The uptake of mediumchain fatty acids lacks negative feedback regulation; therefore, in the absence of MCAD, medium-chain fatty acids accumulate to toxic levels, inducing reactive oxygen species (ROS), mitochondrial damage and failure, and apoptosis. Taken together, our findings uncover a previously unappreciated protective role exerted by MCAD in GBM cells, making it a unique and therapeutically exploitable vulnerability.

cancer biology↗

bioRxiv2020

Diversity Across the Pancreatic Ductal Adenocarcinoma Disease Spectrum Revealed by Network-Anchored Functional Genomics

Cancers are highly complex ecosystems composed of molecularly distinct sub-populations of tumor cells, each exhibiting a unique spectrum of genetic features and phenotypes, and embedded within a complex organ context. To substantially improve clinical outcomes, there is a need to comprehensively define inter- and intra-tumor phenotypic diversity, as well as to understand the genetic dependencies that underlie discrete molecular subpopulations. To this end, we integrated CRISPR-based co-dependency annotations with a tissue-specific co-expression network developed from patient-derived models to establish CoDEX, a framework to quantitatively associate gene-cluster patterns with genetic vulnerabilities in pancreatic ductal adenocarcinoma (PDAC). Using CoDEX, we defined multiple prominent anticorrelated gene-cluster signatures and specific pathway dependencies, both across genetically distinct PDAC models and intratumorally at the single-cell level. Of these, one differential signature recapitulated the characteristics of classical and basal-like PDAC molecular subtypes on a continuous scale. Anchoring genetic dependencies identified through functional genomics within the gene-cluster signature defined fundamental vulnerabilities associated with transcriptomic signatures of PDAC subtypes. Subtype-associated dependencies were validated by feature-barcoded CRISPR knockout of prioritized basal-like-associated genetic vulnerabilities (SMAD4, ILK, and ZEB1) followed by scRNAseq in multiple PDAC models. Silencing of these genes resulted in a significant and directional clonal shift toward the classical-like signature of more indolent tumors. These results validate CoDEX as a novel, quantitative approach to identify specific genetic dependencies within defined molecular contexts that may guide clinical positioning of targeted therapeutics.

cancer biology↗

bioRxiv2020

CREBBP/EP300 mutation is associated with poor outcome in HNSCC and targetable with synthetic cytotoxicity.

Despite radiation forming the curative backbone of over 50% of malignancies, there are no genomically-driven radiation sensitizers for clinical use. We performed in vivo shRNA screening to identify targets generally associated with radiation response as well as those exhibiting a genomic dependency. This identified the histone acetyltransferases CREBBP/EP300 as a target for radiosensitization in combination with radiation in cognate mutant tumors. Further in vitro and in vivo studies confirmed this phenomenon was due to repression of homologous recombination following DNA damage and can be reproduced using chemical inhibition of histone acetyltransferase (HAT), but not bromodomain function. Selected mutations in CREBBP lead to a hyperacetylated state that increases CBP and BRCA1 acetylation, representing a gain of function targets by HAT inhibition. Additionally, mutations in CREBBP/EP300 were associated with recurrence following radiation, in several squamous cell carcinoma cohorts. These findings represent both a novel mechanism of treatment resistance and the potential for genomically-driven treatment.

cancer biology↗

bioRxiv2020

Genetically Encoded, Multivalent Liquid Glycan Array (LiGA)

The Central Dogma of Biology does not allow for the study of glycans using DNA sequencing. We report a "Liquid Glycan Array" (LiGA) platform comprising a library of DNA barcoded M13 virions that display 30-1500 copies of glycans per phage. A LiGA is synthesized by acylation of phage pVIII protein with a dibenzocyclooctyne, followed by ligation of azido-modified glycans. Pulldown of the LiGA with lectins followed by deep sequencing of the barcodes in the bound phage decodes the optimal structure and density of the recognized glycans. The LiGA is target agnostic and can measure the glycan-binding profile of lectins such as CD22 on cells in vitro and immune cells in a live mouse. From a mixture of multivalent glycan probes, LiGAs identifies the glycoconjugates with optimal avidity necessary for binding to lectins on living cells in vitro and in vivo; measurements that cannot be performed with canonical glass slide-based glycan arrays. DedicationThe paper is dedicated to Laura L. Kiessling on the occasion of her 60th birthday.

biochemistry↗

bioRxiv2019

Investigating Coronary Artery Disease methylome through targeted bisulfite sequencing

BackgroundGene environment interactions leading to epigenetic alterations play pivotal role in the pathogenesis of Coronary Artery Disease (CAD). Altered DNA methylation is one such epigenetic factor that could lead to altered disease etiology. In this study, we comprehensively identified methylation sites in several genes that have been previously associated with young CAD patients.\n\nMethodsThe study population consisted of 42 healthy controls and 33 young CAD patients (age group < 50 years). We performed targeted bisulfite sequencing of promoter as well as genic regions of several genes in various pathways like cholesterol synthesis and metabolism, endothelial dysfunction, apoptosis, which are implicated in the development of CAD.\n\nResultsWe observed that the genes like GALNT2, HMGCR were hypermethylated in the promoter whereas LDLR gene promoter was hypomethylated indicating that intracellular LDL uptake was higher in CAD patients. Although APOA1 did not show significant change in methylation but APOC3 and APOA5 showed variation in methylation in promoter and exonic regions. Glucokinase (GCK) and endothelial nitric oxide synthase 3 (NOS3) were hyper methylated in the promoter. Genes involved in apoptosis (BAX/BCL2/AKT2) and inflammation (PHACTR1/LCK) also showed differential methylation between controls and CAD patients.\n\nConclusionsThis study is unique because it highlights important gene methylation alterations which might predict the risk of young CAD in Indian population. Large scale studies in different populations would be important for validating our findings and understanding the epigenetic events associated with CAD.

genomics↗

bioRxiv2018

Mapping transcriptionally equivalent populations across single cell RNA-seq datasets

The power of single cell RNA sequencing (scRNA-seq) stems from its ability to uncover cell type-dependent phenotypes, which rests on the accuracy of cell type identification. However, resolving cell types within and, thus, comparison of scRNA-seq data across conditions is challenging due to technical factors such as sparsity, low number of cells and batch effect. To address these challenges we developed scID (Single Cell IDentification), which uses the framework of Fishers Linear Discriminant Analysis to identify transcriptionally related cell types between scRNA-seq datasets. We demonstrate the accuracy and performance of scID relative to existing methods on several published datasets. By increasing power to identify transcriptionally similar cell types across datasets, scID enhances investigators ability to extract biological insights from scRNA-seq data.

bioinformatics↗

bioRxiv2018

Process-specific somatic mutation distributions vary with three-dimensional genome structure

Somatic mutations arise during the life history of a cell. Mutations occurring in cancer driver genes may ultimately lead to the development of clinically detectable disease. Nascent cancer lineages continue to acquire somatic mutations throughout the neoplastic process and during cancer evolution (Martincorena and Campbell, 2015). Extrinsic and endogenous mutagenic factors contribute to the accumulation of these somatic mutations (Zhang and Pellman, 2015). Understanding the underlying factors generating somatic mutations is crucial for developing potential preventive, therapeutic and clinical decisions. Earlier studies have revealed that DNA replication timing (Stamatoyannopoulos et al., 2009) and chromatin modifications (Schuster-Bockler and Lehner, 2012) are associated with variations in mutational density. What is unclear from these early studies, however, is whether all extrinsic and exogenous factors that drive somatic mutational processes share a similar relationship with chromatin state and structure. In order to understand the interplay between spatial genome organization and specific individual mutational processes, we report here a study of 3000 tumor-normal pair whole genome datasets from more than 40 different human cancer types. Our analyses revealed that different mutational processes lead to distinct somatic mutation distributions between chromatin folding domains. APOBEC- or MSI-related mutations are enriched in transcriptionally-active domains while mutations occurring due to tobacco-smoke, ultraviolet (UV) light exposure or a signature of unknown aetiology (signature 17) enrich predominantly in transcriptionally-inactive domains. Active mutational processes dictate the mutation distributions in cancer genomes, and we show that mutational distributions shift during cancer evolution upon mutational processes switch. Moreover, a dramatic instance of extreme chromatin structure in humans, that of the unique folding pattern of the inactive X-chromosome leads to distinct somatic mutation distribution on X chromosome in females compared to males in various cancer types. Overall, the interplay between three-dimensional genome organization and active mutational processes has a substantial influence on the large-scale mutation rate variations observed in human cancer.

genomics↗

bioRxiv2017

Evaluation and Design of Genome-wide CRISPR/Cas9 Knockout Screens

The adaptation of CRISPR/Cas9 technology to mammalian cell lines is transforming the study of human functional genomics. Pooled libraries of CRISPR guide RNAs (gRNAs), targeting human protein-coding genes and encoded in viral vectors, have been used to systematically create gene knockouts in a variety of human cancer and immortalized cell lines, in an effort to identify whether these knockouts cause cellular fitness defects. Previous work has shown that CRISPR screens are more sensitive and specific than pooled library shRNA screens in similar assays, but currently there exists significant variability across CRISPR library designs and experimental protocols. In this study, we re-analyze 17 genome-scale knockout screens in human cell lines from three research groups using three different genome-scale gRNA libraries, using the Bayesian Analysis of Gene Essentiality (BAGEL) algorithm to identify essential genes, to refine and expand our previously defined set of human core essential genes, from 360 to 684 genes. We use this expanded set of reference Core Essential Genes (CEG2), plus empirical data from six CRISPR knockout screens, to guide the design of a sequence-optimized gRNA library, the Toronto KnockOut version 3.0 (TKOv3) library. We demonstrate the high effectiveness of the library relative to reference sets of essential and nonessential genes as well as other screens using similar approaches. The optimized TKOv3 library, combined with the CEG2 reference set, provide an efficient, highly optimized platform for performing and assessing gene knockout screens in human cell lines.

systems biology↗

bioRxiv2016

Extrachromosal DNA elements can drive disease evolution in glioblastoma

To understand how genomic heterogeneity of glioblastoma (GBM) contributes to the poor response to therapy, which is characteristic of this disease, we performed DNA and RNA sequencing on GBM tumor samples and the neurospheres and orthotopic xenograft models derived from them. We used the resulting data set to show that somatic driver alterations including single nucleotide variants, focal DNA alterations, and oncogene amplification in extrachromosomal DNA (ecDNA) elements were in majority propagated from tumor to model systems. In several instances, ecDNAs and chromosomal alterations demonstrated divergent inheritance patterns and clonal selection dynamics during cell culture and xenografting. Longitudinal patient tumor profiling showed that oncogenic ecDNAs are frequently retained after disease recurrence. Our analysis shows that extrachromosomal elements increase the genomic heterogeneity during tumor evolution of glioblastoma, independent of chromosomal DNA alterations.

cancer biology↗